Presentation is loading. Please wait.

Presentation is loading. Please wait.

Ashok Agarwal 1 BaBar MC Production on the Canadian Grid using a Web Services Approach Ashok Agarwal, Ron Desmarais, Ian Gable, Sergey Popov, Sydney Schaffer,

Similar presentations


Presentation on theme: "Ashok Agarwal 1 BaBar MC Production on the Canadian Grid using a Web Services Approach Ashok Agarwal, Ron Desmarais, Ian Gable, Sergey Popov, Sydney Schaffer,"— Presentation transcript:

1 Ashok Agarwal 1 BaBar MC Production on the Canadian Grid using a Web Services Approach Ashok Agarwal, Ron Desmarais, Ian Gable, Sergey Popov, Sydney Schaffer, Cameron Sobie, Randall Sobie, Tristan Sulivan, Daniel Vanderster University of Victoria

2 Ashok Agarwal 2 Overview of GridX1 – A GT2 Grid Use Canadian resources calliope, mercury, mcgill Clusters: standard Globus Toolkit 2 (GT2) Resource Mgmt: CondorG-based MS Condor Brokering MyProxy credential repo Central monitoring and accounting with web GUI

3 Ashok Agarwal 3 Limitations of GridX1 Due to numerous service-specific protocols e.g. GRAM, MyProxy, Condor 1.Difficult to extend: Adding a new service involves modifying protocol, or developing a new one 2.Compatibility issues: Lack of protocol standardization Backwards compatibility is not perfect (protocols modified between releases) 3.Firewall problems: Each service uses its own TCP port Many ports must be opened by each institution this may conflict with local policies 4.Security vulnerabilities: The GRAM job service runs as root, which could lead to a compromised resource Access is often limited to trusted hosts, limiting usefulness of the service Solution: Web Services Resource Framework (WSRF) Globus Toolkit v.4

4 Ashok Agarwal 4 Globus Toolkit v.4 Advantages WSRF solves the 4 key problems with GridX1 1.Easy to extend One common protocol (SOAP) Easy to develop new WSRF services 2.Seamless upgrade support Changes to service interfaces are described in WSDL 3.Reduced firewall problems Fewer ports (the service container) Non-privileged ports 4.Good security Service container runs as non-privileged user

5 Ashok Agarwal 5 WSRF-Based Grid Consists of multiple metaschedulers Central resource registry to store the resource attributes, RFT and LRMS Having multiple metaschedulers and registries gives high scalability and reliability of the grid Proposed WSRF-based grid

6 Ashok Agarwal 6 Important Features Condor-G used as the metascheduler Automatic registering of resource ClassAds to the central registry Automatic ClassAds extraction from the registry to the metascheduler for matchmaking Incorporation of input/output file staging Job submission using WS-GRAM or Condor_Submit Web-based monitoring Metascheduler Service

7 Ashok Agarwal 7 Publishing Tool An information provider script runs on every Grid resource and generates the resource ClassAd in GLUE 1.2 scheme Inserts the resource information in the form of XML-formatted Condor ClassAds into the local WS-MDS Publishes the resource information into the WS-MDS of the central registry ClassAd Publishing Tool

8 Ashok Agarwal 8 ClassAd Extractor Java application software Runs periodically on the metascheduler Extracts compute resources ClassAds in GLUE 1.2 scheme Converts the XML data into the Condor ClassAds for each resource Publishes these ClassAds to the Condor collector for jobs matchmaking MdsClassAdExtractor Tool

9 Ashok Agarwal 9 BaBar MC Grid Requirements Metascheduler head node –Install Condor-G and GT4 –Install BaBar software –Set up metascheduler –Set up ClassAd extraction tool Grid Resources Head Node –Set up Portable Batch System (PBS) to act as the local resource management system –Install BaBar software –Set up classad.pm to advertise the local resource information –Set up ClassAd publishing tool

10 Ashok Agarwal 10 BaBar MC Production Setup Using Resource Broker

11 Ashok Agarwal 11 Performance Test % Eff = CPU Time / Wall Time Comparison of CPU time and percent efficiency (% Eff)

12 Ashok Agarwal 12 JSAM: Job Submission Tool

13 Ashok Agarwal 13 GridX1 Condor-G Monitoring

14 Ashok Agarwal 14 WSRF-Based Grid Monitoring

15 Ashok Agarwal 15 Conclusion With the WSRF (GT4), we have developed A metascheduling service using Condor-G Resource information provider Automatic ClassAd extraction tool Job submission client tool Execution of BaBar jobs is successful on the GT4 grid Web-based monitoring is useful for providing the status of grid resources and the jobs Monitoring is based on Condor_history. Work is in progress to improve monitoring using condor_quill. Production will start soon on this WSRF-based grid


Download ppt "Ashok Agarwal 1 BaBar MC Production on the Canadian Grid using a Web Services Approach Ashok Agarwal, Ron Desmarais, Ian Gable, Sergey Popov, Sydney Schaffer,"

Similar presentations


Ads by Google